Skip to content

P104 leak corrections: third-engine check of #6067/#6069 on PR6030, and a scalar-level witness on PR5907 (no new κ) - #6108

Open
JosieElliston wants to merge 2 commits into
CrocSwap:mainfrom
JosieElliston:s41-p104-post-cut-linear
Open

JosieElliston wants to merge 2 commits into
CrocSwap:mainfrom
JosieElliston:s41-p104-post-cut-linear

Conversation

@JosieElliston

Copy link
Copy Markdown
Contributor

Result

Scoped exact result, no new κ. Selected record unchanged (PR6030, κ 0.003125804, main 721c03822).

Priority first. On the selected PR6030 word, the questions PR6045 left open were answered first: combination
carriers in #6067 (apoursh), and rewritten or arbitrary post-cut edits in #6069 (jon314159). Both report the
4,033 PR5718-shape leak classes as 4,009 information-lost, 22 with no chart-local combination carrier, and 2
(cuts 13637/13639, helper 572) erased. I built this package without knowing about them and found the overlap only at
submission. So it is offered as:

  1. A third independent engine for Exclude combination-carrier correction programs on the selected P104 word (no new κ) #6067/P104: certify fixed-word carrier and endpoint exclusions (no new κ) #6069. It is an exact left-kernel tracker (29 vectors per state,
    audited at every leak-class cut against the exact field matrix, plus 300 brute-force small words). It gives the
    same 4,009 / 22 / 2. Its 24 residual classes match P104: certify fixed-word carrier and endpoint exclusions (no new κ) #6069's global_screen.survivors class for class (local
    comparison).
  2. New: the PR5907 word, where the exclusion fails. 4,180 of 4,206 classes are information-lost, 23 have no
    usable combination, and 2 are erased. One class is legalized at the scalar level: cut after 13776, chart 1031,
    helper 699, leak to role 149. The seven-field combination 696 − 698 + 699 − 700 − 702 − 767 + 792 is zero in
    the original word and equals z − ℓ in the split word. Reading it into 697, 698 and 700 directly after the cut
    lowers to 18 ADDs, 1 SCALE and 1 COPY, all chart-local. A plain simulator shows every outside-V output equals the
    original exactly, and z reaches roles 698, 700 and 702. PR6040's pairing test passes, and three perturbed
    programs are rejected. So the PR6030 exclusion depends on the word, and the combination model is not vacuous on
    real data. This is not a split proposal: native maps (including the overwrite COPY), lifetimes, cost and the
    caller are unchecked.

Package: research/p104-post-cut-information-loss/ only.

Reproduce

python3.12 -B research/p104-post-cut-information-loss/verify.py          # PASS_P104_POST_CUT_INFORMATION_LOSS, ~17 s
python3.12 -B scripts/research_package.py --package research/p104-post-cut-information-loss
python3.12 -B scripts/check_research_commit.py --package research/p104-post-cut-information-loss

Passed locally on macOS, CPython 3.12.14, in normal mode and in -I -W error -O mode, plus the package-metadata and
commit-bytes checks. Codex ran a read-only adversarial review. It found no blocking issue and two minors: the audit
scope wording, now "every leak-class cut", and the physical legality of the COPY, now stated as unchecked. Hosted CI
was not consulted.

Scope and remaining obligations

Credit

apoursh (#6067) and jon314159 (#6069) published the PR6030 exclusion and the information-collision argument first.
Also credited: Nick Petalas (#6040 census, pinned unmodified), Matarener (#5994), flyingviper (#5718), Ali Romman
(#5599), DreamingOfClouds (#5907/#6030 words), and #6045 (leakfix.py, pinned unmodified). Prepared for
JosieElliston by Claude Opus 5.5 (Anthropic) via Claude Code, as one session of an unattended overnight run. No human
review. Apache-2.0.

🤖 Generated with Claude Code

JosieElliston and others added 2 commits October 28, 2026 04:01
…07 witness (no new κ)

Left-kernel tracker reproduces PR6067/PR6069 on the selected PR6030 word
(4,009 information-lost, 22 no chart-local combination, 2 erased); on PR5907
one class is legalized at the scalar level by a 7-field chart-local
combination program, checked by direct simulation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant